Mercurial > pylearn
comparison doc/v2_planning/datalearn.txt @ 1364:01157763c2d7
Reply to Razvan
| author | Olivier Delalleau <delallea@iro> |
|---|---|
| date | Fri, 12 Nov 2010 11:36:30 -0500 |
| parents | 18b2ebec6bca |
| children | 049b99f4b323 |
comparison
equal
deleted
inserted
replaced
| 1363:18b2ebec6bca | 1364:01157763c2d7 |
|---|---|
| 217 the pipeline logic is separated from the data, so you can use the same | 217 the pipeline logic is separated from the data, so you can use the same |
| 218 transformation with different data easily, while in (1) you write the | 218 transformation with different data easily, while in (1) you write the |
| 219 transformation rooted in a dataset, and if you want same transformation | 219 transformation rooted in a dataset, and if you want same transformation |
| 220 for a different dataset you have to re-write everything. | 220 for a different dataset you have to re-write everything. |
| 221 | 221 |
| 222 OD replies: Still not sure I understand. If you have a "graph" function that | |
| 223 takes a dataset as input and outputs a new dataset, you can use this same | |
| 224 function with both (1) and (2). With (2) it is: | |
| 225 theano.function([index], graph(my_dataset)[index].variable) | |
| 226 while with (1) the same function is compiled implicitly with: | |
| 227 for sample in graph(my_dataset): | |
| 228 ... | |
| 222 | 229 |
| 223 - in approach (1) the initial dataset object (the one that loads the data) | 230 - in approach (1) the initial dataset object (the one that loads the data) |
| 224 decides if you will use shared variables and indices to deal with the | 231 decides if you will use shared variables and indices to deal with the |
| 225 dataset or if you will use ``theano.tensor.matrix`` and not the user( at | 232 dataset or if you will use ``theano.tensor.matrix`` and not the user( at |
| 226 least not without hacking the code). Of course whoever writes that class | 233 least not without hacking the code). Of course whoever writes that class |